How we could do TEN TIMES more covid tests, for free
HTML-код
- Опубликовано: 6 июл 2024
- We need to do more testing for coronavirus. What if we could do ten times more tests, without needing any more test kits, just using maths?
This is such an important idea, and it’s being woefully underused by countries around the world. Please share this video-the more widely this simple, elegant way to fight covid is known, the better!
Video chapters
00:00 Introduction
00:27 PCR testing
01:06 Simple pooling
04:13 2D pooling
06:43 Kirkman’s schoolgirl problem
09:10 Kirkman’s schoolgirls’ coronavirus problem
10:46 Cleverer combinatorial methods (P-BEST)
12:18 Let’s test EVERYONE
13:45 WHY IS NO-ONE DOING THIS?!?!
If you want to know more about why testing is to critical to controlling coronavirus, check out the section on test, trace and isolate in my epic covideo from April: • Coronavirus: what the ...
Sources
Here are some of the sources I used when researching this video:
Preprint on 2D pooling using 96- and 384-well plates which got me excited in March: www.medrxiv.org/content/10.11...
Article by one of the Rwandan mathematicians who worked on hypercube testing (though it doesn’t explicitly mention high-dimensional pooling): theconversation.com/rwandas-c...
HOT OFF THE PRESS on October 21st, the hypercube-based pooling method in Nature! www.nature.com/articles/s4158...
Kirkman’s schoolgirl problem: en.wikipedia.org/wiki/Kirkman...
You can read more about how to solve it here: cs.lmu.edu/~ray/notes/kirkman/
And you can find a cool visualisation of the solution here: math.stackexchange.com/a/1204049
Girls’ names selected at random from the top 10 names for each letter between 1996 and 2018 in the UK: www.ons.gov.uk/peoplepopulati...
Article on the P-BEST algorithm: www.nytimes.com/2020/08/21/he...
The P-BEST paper: advances.sciencemag.org/conte...
Paul Romer discusses his mass testing plan in this interview with the brilliant @Healthcare Triage (whose whole channel is a great source of coronavirus information): • Paul Romer's Coronavir...
It’s also detailed on his website roadmap.paulromer.net/ and in this written interview www.newyorker.com/news/q-and-...
Wikipedia list of places around the world using pooled testing for coronavirus: en.wikipedia.org/wiki/List_of...
Coronavirus testing data via the excellent Our World in Data: ourworldindata.org/grapher/fu... (total of 512,080,611 tests retrieved on 30/09/2020…sorry, it took me a long time to finish this video!)
Interesting things that didn't make it into the video
A great article on how this is the exact same maths as the game of Dobble (aka Spot It! in the US) puzzlewocky.com/games/the-mat...
Combinatorial testing without needing robots, using punched cards www.smarterbetter.design/orig...
Images
Nasopharyngeal swab technique diagram via www.cdc.gov/vaccines/pubs/sur...
384-well plate image via www.flickr.com/photos/6486047...
The pretty Victorian church is Ash Church in Kent, from a book published in 1864 www.flickr.com/photos/british...
Errata
Obviously where I mention Kirkman triples for ‘many different combinations of total numbers of students and group sizes…’ around 08:05 mark where you see the asterisk…well, triples means a group size of three, doesn’t it? I only noticed this obvious mistake in the edit! Also, the solution was first published in 1968. Not the best fact-checked paragraph in the script…
Thanks to trifonTAF for pointing out in the comments that I slipped up describing the P-BEST algorithm: it divides samples into 48 pools *of 48 samples each*, and each sample appears in 6 pools. D'oh! 
Наука
Well, this blew up (thanks, Steve Mould!) so I’ve not got time to reply to every comment individually, but I thought I’d stick a quick FAQ up here as a few things keep coming up!
*How does this actually work in practice?*
When the swabs arrive at the lab, they get dunked into some liquid to extract the RNA. (No further questions, I’m a _computational_ biologist! ;) ) You can then pipette small amounts of this liquid into wells to mix them together, and/or keep a small amount of it back to do a second test later if needed.
We don’t need to do multiple swabs, thus delaying matters, and we definitely don’t need to stick a potentially coronavirus-y swab up multiple people’s noses!!
*Does this affect test accuracy?*
I thought this was too complicated for an already-long video, but you asked. :) The short answer is: probably not!
The obvious worry is false negatives, because you’re mixing samples together and diluting them so maybe you’d miss cases. I read a few papers which went into detail on this, including this one: advances.sciencemag.org/content/6/37/eabc5961 It's for the P-BEST algorithm, and they check the test sensitivity with real PCR tests, and conclude it doesn't affect false negatives too badly.
Another problem is false positives. However, combinatorial testing should reduce false positives: imagine you got a false positive in one pool, you might then have (say) three other pools containing each of the samples in the positive one to clear their name! Depending on the biological details you might then want to re-test those samples just to be sure, but either way I think it should reduce the false positive rate.
*Why aren’t we doing this?*
I don’t know. There are some logistical issues, but unless they make things ten times worse in the lab we’re still winning!
*Update:* With fantastic timing, the Rwandan group's paper on hypercube-based test pooling is published in Nature today! www.nature.com/articles/s41586-020-2885-5
Thanks for watching, and if you’ve got any more questions that aren’t covered above feel free to stick them here and I’ll try to answer!
Woh! Woh! Woh! Woh! Woh! Woh! Woh there buddy!!! Don't you start using "logic" on us now! We've done just fine using our alternative logic thank you very much! Besides, where ya gonna get all these "Mathsy" type people anyways? We don't hardly have enough of them here in the states to make a dent in any of our processes. So there! (Okay, this was sarcasm.)
These ideas are awesome. And yeah, they could do well, but really, it has to also be followed up with a complete no brainer way of ensuring the average person conducts the procedures correctly. They won't understand why they are doing it, so making silly mistakes will be inevitable. Yes, it COULD work, in theory. But humans are at every facet of this approach. To factor the error potential out may very well be impossible. Perhaps if we had robots doing human jobs, not just in a lab, but all up and down the work environment, yes, then maybe that will work. But we're not exactly there yet. Robots are barely faking it at the moment. (See recent robot conventions on youtube, not too impressive [for bridging the gap to replace humans that is]) Should we not try? Heck, sure why not? I'm not sure we could make it much worse than it already is (in the USA at least, France and Italy aren't looking so hot either).
I can't help but feel that maybe we are lacking tools to make this easy enough for the layman. We can't assume that swab testers have the knowledge to perform combinatorics math reliably. Even though it does sound like they should know how...
Steve Mould sent me here.
(He's a wise man.)
He really is. Everything I know about Pythagorean siphons I learned from him.
Same
Same
I've never looked at my flush the same again
amazing! found your video through steve mould
yep
yep
yep
Yep
yep
Thanks Steve Mould 👍
I agree! What a guy!
I got really emotional watching this. It's so nice to listen to someone who's not a complete bumbling obnoxious plank talk about Corana Virus for 15 minutes. Great Stuff!
He starts out with "the coronavirus pandemic" as if it was real. Sorry can't take him seriously.
@@StefanReich lets just hope you are joking.
As an analytical chemist, one reason pooled testing isn't used in most instances is because of two factors:
1) false positives. If you have a false positive in a situation where you are testing 1:1, you only need to retest that one sample twice to correctly identify the false positive.
If a pooled result comes back as a false positive, you would then have to test all the individuals in the pool twice (and possible retest the pool as well) to comfortably confirm it was a false positive.
2) capacity. If you are pooling tests, and at maximum capacity, and then you get (either by chance or by rates going up) an increase in positives, you then need to do a lot of extra testing you don't necessarily have capacity for, meaning the lab becomes overwhelmed (which is the worst thing to happen as positive rates increase). 1:1 testing more or less assures this can never happen
This all makes sense but I think some out of the box thinking here may be helpful. That is to say *completely supplementary* testing. Right now we only have the labs and testing for people who exhibit symptoms or have some high-risk which means regular testing.
If we spent the money to add lab capacity to do *this* kind of testing then we could do 'test everybody' type regime that could help detect hot-spots very quickly and give groups some indication of risk.
And would that be enough to open up? ... Well, probably not. The lag time is probably still too great and the ability for quick spread too high for this to be enough as the virus is spreading quickly -- but maybe back in the summer months it would have been just the thing to really stamp out the virus so there were fewer active cases going into late fall and winter.
So basically, I talked myself out of this being a solution now (for this pandemic anyway, on the assumption we have a widespread vaccine in April/May). Still fascinating math! :)
This is how the University of Cambridge is managing to provide weekly asymptomatic tests to students. Apparently experts seem to be pretty good at doing stuff
Love this I wish this wasn't all so politicized and we took a more mathematical approach
@Jitwitdastick JRE III yeah? In my country it is the other side who bought a shady Chinese vaccine with taxpayer money without any kind of auditing and now wants to use the police to force people to take it. Suddenly it is your body, their rules.
@@mariobros7834 sure thing. I'm sure that's happening
It's hard not to politicize it when there are groups that don't believe in math.
Germany does exactly this
You may be interested to know that the University of Cambridge is using a pooling method for its weekly testing and these pools are comprised of households (people who share facilities so are counted as in the same bubble like a family which shares a house) which means that positive test results are likely to be grouped together further increasing efficiency.
For those who want to know more they actually group the tests on a more fundamental level than in this video as we all put our swabs into the same testing tube (I am not currently familiar with exactly what happens when a test comes back positive but it may well be the case that they send out a new set of tests to be done from scratch again - this extra delay not being a problem as if someone in your household is positive everyone needs to isolate anyway)
@@antisocialbob968 wait so everyone gets swabbed individually but the samples get tested as one?
@@notme-ji5uo yes all the individual swabs which are from each of the (up to 10) people get put into the same testing tube so I don't see how they could separate out who is who
Found this on Twitter and regardless of the subject matter it’s the first time I’ve found maths remotely interesting since (or more likely because of) school. Perfect choice of T-shirt too! 👏🏼💪🏼
I was trying to figure out what the shirt represents, where's it from?
@@samrusoff it’s a moon phases t-shirt but it’s also very similar to the 91 sample tray mention in the vid. No idea where he got it from though, sorry.
Also. If the pools are “contact groups”, for example , families ,work groups or passengers, etc. . you probably don’t need to test individually everytime. Since you will need to isolate all of them anyway in case they have the virus but it’s still not enough for the PCR to detect. In that case. You can treat all the group as positive an test another “layer” of contacts of all the members of that group in order to try to find new possible cases sooner.
Please almighty RUclips algorithm, let this go viral. People making decisions need to know this
Great video! Nice graphics, really easy to follow. Good on ya, mate.
2D pooling also reduces false negatives because every sample gets tested at least twice. This is awesome!
This was great, subbed! Hi from Steve Mould!!
Amazing Video! Math should rule the world!
P.S Thanks for "Video chapters".
Awesome info. Thank you
Just got this in my RUclips recommendation. Godspeed, brother.
One of the most underrated channels out there with video production better than that of channels with 1M+ subscribers!
Proud to say I was watching you before you inevitably blow up Andrew!
Haha thank you! Hope you're right :)
I'm new to the channel and had to double check for a moment that the subscriber count is actually 1,6K and not 1,5M
The pool method of testing has long been implemented when giving blood in Belgium. This also gives you the option to decide yourself whether you want an extra (for example hiv) test when you get a letter that your pool was found positive.
Brilliantly explained. And fascinating. Send to Boris Johnson immediately!!
Thanks Steve Mould for the heads up in this!!
Nice video, really well put together
Thank you!
Great up n coming channel it would seem, thanks Steve Mould for the recommendation
A joy to consider such mathtastic measures, thanks
Steve Mould sent me here too. Very well done !!! Thanks !!!
Love this! Excellent!
Very well explained!
Simple and clear. Great explanation, you're very talented at making informative content! I was very surprised seeing at the end of the video how few views it has. I hope your channel grows (with a little help from Steve Mould)!
Thanks! And always happy to have a leg up from the mighty Mould!
@@DrAndrewSteele he is one of the nice guys of RUclips! Well, I did my part by subscribing! Off to check your other videos!
@@tommyvega7948 Thanks dude, hope you enjoy!
Brilliant. Why I didn't think of this before. It is so obvious.
Liked, subscribed, thank you for pointing this out, and thanks Steve Mould for pointing you out.
I like this video. Thanks. I subscribed.
This needs to be seen by more people
Great video, more visual aids for the situations you're talking about please. Even simple visual aids
I'll see what I can do! This was a really hard one to animate in places but I am definitely a fan of using visual aids where possible-otherwise, why not just make a podcast? :)
This is a bit like error correction in coding theory. Very interesting!
That was a very interesting video! I dig it
Brilliant!
Really great explanation. It is frustrating to think of the usefulness of such strategies, especially the more simple to understand ones, aren't being used more prevalently. We can only hope this changes.
Hi, in Australia this is how we've been doing pcr testing since march
its the same algorithm of an led screen, how we control each pixel.
I'm not usually one to leave RUclips comments but I just wanted to say this video was great! Just subscribed and look forward to more of the same, and yes Steve sent me :p
I don't know what pooling strategies are used here in Sweden, but I do know pooling is used.
Could this be done for testing a sample size of 200 i.e. a commercial flight?
Great video!
This sounds like a spreadsheet I made once.
Teachers have to do supervision of students at lunch and recess. The problem is that the teachers hate giving up their break time. There are a total of 10 areas that need supervision sometimes at the same time. The teachers don't want to do supervision 2 days in a row, or the same area more than once in a row and certainly not twice in the same day. They also can't do certain areas or times due to schedule conflicts or other factors. There could also be prefences to avoid certain areas but not a hard no.
You also need to make sure that nobody does more or less than their fair share so if a few had to do more than average one month they will do less in the next month to balance it out.
Oh yeah and not all teachers are full time. Some are half time and one is .6 time. Let me know if you want to know how I solved this.
From what I understand, one issue with such massive testing is how to collect the test samples. There are already often queues in front of testing facilities and people have to wait to get tested. Nevertheless, I absolutely love the presented ideas!
This is clever Idea but the consistency of the sample contains this RNA will be hard to find. Maybe ?
Awesome video! What happens when we factor in false negative and false positive rates?
Clicked on the video only because I own your shirt!
Its a good shirt!
Struggling to understand why this isn't implemented... It's very frustrating
How robust are these methods when considering how false positives/negatives will propagate?
Its the only thing I can imagine being a problem. I know the PCR test is meant to have very high sensitivity and specificity but the values are always going to be imprecisely known given how cross contamination is so unpredictable. I still hear the argument it's best to avoid testing populations expected to have very low prevalence of covid due to the whole Bayesian problem where number of false positives becomes significant compared to true positives. But it should be possible to put an upper bound on the specificity of the PCR test so far at least given the lack correlation between tests administered and number of positives cases.
Yeah, it really is. It's taken me a while to get round to making this video too-people have been going on about it for months!
Sorry to copy/paste a reply but this from another comment hopefully answers your question: You're right that it makes things more complicated-too complicated for my video, I decided! :) I read a few papers which went into detail on this, including this one: advances.sciencemag.org/content/6/37/eabc5961 It's for the P-BEST algorithm, and they actually check sensitivity with real PCR tests, and conclude it doesn't affect false negatives too badly. I think combinatorial testing should reduce false positives, as you'd have multiple other pools containing any given sample to clear their good name as it were!
this isn't implemented because it creates a very difficult workflow to keep track of pools and test groups. it's a very good idea on paper that translates poorly into a real lab environment.
other stuff:
a glaring issue of using % of positive tests is that many patients are tested multiple times due to false positives/confirmation testing, and even duplicate testing.
false positives can be caused by similar DNA/RNA sequences that are similar to COVID's. PCR errors are sometimes unpredictable and uncontrollable due to transcription errors made by the enzymes used in the PCR testing.
tests in the medical field (in the USA) are run with 95% confidence intervals, meaning they are reliable 95% of the time, verified by QC every 24 hours. if the test cannot meet this criteria, it is not suitable for mass-scale testing.
i work with the one of the major Alinity instruments used for mass testing of COVID samples in USA.
While this Video and the work behind it is great, it is problematic that the reasons why this way of testing is not feasible in a lab environment are just brushed over. As stated by the comment above, the work flow of this is too complicated and "just using robots" makes this whole issue seem like everything we need is available but politicians are just too blind/lazy to implement this. One of my friends set up and operates one of the "usual" testing robots - a highly complicated matter and it does not always run smoothly. Get this unsmooth operation, false positives and negatives, etc. into one pot and you're headed for disaster.
One thing mathematicians and also physicists sometimes just brush over in their proposals is if their theories can actually be put to work as easily as they think (which usually isn't the case). This video frustrates people that have no background in how the testing works, what further problems come with this proposed method and leaves them feeling quite right in their opinion that our governments just suck/are dumb.
@@Konfusette Thanks for the comments, Nico and Aaron! I freely admit I've not gone into detail here (and I'm a dreaded computational biologist who's never used a pipette), but smarter wet lab people than me think this is viable and I trust them! The logistics aren't totally straightforward, sure, but unless they make testing ten times harder you're still winning!
The robots remark was a bit of a throwaway. :) One more practical idea for humans I've seen are these 'origami assays' which use printed and punched cue cards to guide where to drip samples onto a plate for combinatorial testing, see www.smarterbetter.design/origamiassays/default/instructions I did have a mention of this in the original script but it was just getting too long!
@@Konfusette not only is the operation not very smooth, but there is a massive shortage of Clinical Laboratory Scientists right now. there's a 20-30 year age gap in the work force. if nurses and doctors are the stars of the show in medicine, clinical laboratory scientists are the stage crew. no screen time, but the show wouldn't work without them.
This was actually... Really good wow
This reminds me of the 1000 bottles of wine riddle
What about Hamming code for multiple error detection that could possibly lower the number even more.
Great video, sound needs to have more treble and a bit of reverb.
That strategy is implemented in Denmark. We're pooling three samples at a time, but that is more in order to reduce spent test kits. Worst effect is that a person who is positive gets their result later.
Nice video. Headline: Block Designs allow efficient characteristic evaluation shocker... who'd have thunk combinatorics had any practical use (except for Marketeering people for the last 50 years)
Brilliant
How would these methods deal with false negatives? Would the overall false-negative-rate increase?
What about using a binary mapping?
THIS IS SO COOL! It's incredibly frustrating that this isn't being done. All of the people complaining that they want to go get a haircut and go to the movies could do that if we tested everyone! Imagine! We could go back to normal!
Surely you can do some kind of binary search here? That would only require O(k*log(n)) tests for pools of size n, with k positive cases.
Is it group testing? I just wanna guess. I'm excited.
Quite suffisticated method! But still, there is the problem at this stage that we are still trying to build models of the propagation of the virus, so in some sense, the measure for this method is also the target of the method, wich makes it quite conflicting as a solution to the problem. Since the actual ratio of true positives to true negatives is rapidly changing, depending on the times, how would you decide when to change the measure without affecting the theoretic model of propagation?
How would any individual test error resulting in, false positives, and false negatives, affect the maths?
I don't think it would effect it that bad
Interesting stuff. Steve sad Hi. Subbed.
i work with one of the Alinity machines (the high-volume covid analyzers) we tried pooling samples at one point, but due to the volume of tests, the limitations of the instruments, lack of fully-automated workflow, and the hospital's treatment/isolation schedule being dependent on the turn around time of these tests, we've found pooling samples to be troublesome and inefficient.
surely the system isnt build around pooling...
@@klausbrinck2137 that is correct in the majority of cases. that being said, some companies/systems are willing to/already have re-formatted their workflow to run in such a way. of course, changing so dramatically costs money and more importantly time. it may also require retraining of staff.
Great idea/ But would it not make sense to use hexadecimal numbers?
Test 16 (or 32/64 etc...)
If negative - clear.
If positive - split - test 8.
If negative, clear.
Test 4.
If negative, clear.
In the case of positive - split and test half.
Repeat.
This way you could combine any number of samples and eliminate any clear group.
Any positive group would be split and retested.
Watched this several hours ago and I'm still trying to decipher the t-shirt.
Me too! At first thought it was something to do with phases of the moon but I’m not so sure...
@@TSutton Solar Eclipse phases..
@@kukunishad Haha, all good suggestions! I too have spent lots of time trying to work out what it means…
@@DrAndrewSteeleReally?? I thought you like astronomy so you wore that tshirt :D
@@kukunishad Looking ambiguously astronomical was definitely a big factor in its purchase!
The limiting factors will end up being swabs and PPE if this is used as a broad test of the population. You can get around the PPE by having people swab themselves under supervision. Not sure about the swabs if we test very large portions of the population regularly though.
Where is the limiting factor in testing at the moment? Is it the lab space and biochemical analysis or is it the swabs and the kits that get sent out to people?
I don't know the details and it probably varies a bit by country, but I think in the UK it's at least partly due to lab capacity-I don't know if you've been listening to More or Less's excellent coverage on this, but the Government's 100,000 test target in April was met by counting swabs that were posted on that day, but many were of course never returned! So that suggests to me that we're not short on swabs…
Not once did you take into account dilution factor from multiple samples and the need to run the PCR for a higher cycle count.
The only problem with this calculations is that it totally ignores the fact the lab testing is only one part of the problem , the real bottleneck timewise is the nasal sample collection as people must be handled with significant time and space separation to avoid contamination.
BRILLIANT.
Basically the same idea as hamming codes
I thought about something similar when testing blood samples from donation for infected samples (HIV e.g.). Since I'm Software Engineer, I am familiar with binary search and know its advantages.
I've heard of some of these ideas before but not fully had them explained so interesting to hear a bit more detail. However, some of these ideas require diluting the sample and I wonder how much this will affect accurately identifying actual cases.
PCR doesn't give a binary answer. PCR involves repeated cycles to amplify the viral genetic material. If the viral RNA is detected within a set number of cycles the sample is said to be positive. If your sample is divided (e.g. one for pooling and the other for confirmation or divided into two separate groups) arguably that would require more cycles to detect the virus and then actually increase the chance of a false positive due to background noise.
This isn't as much of a problem as it seems-see the FAQ in my pinned comment!
Tests are done in groups using this method in China.
In theory you could use even less test as say a group of 8 people test Positive then test in 2 groups for each half then test the 2 remaining and and then you only used 4 tests in a senerio that you describe would use 9 the problem is it would increase the time for results exponentially since you would have to have the larger group tested before the smaller ones it would take more time so the real problem is which you want to value number of tests or time to get results
So why not? What's the limiting factor elsewhere in the system?
Getting the swabs done? Logistics of transporting the materials or samples?
I don't know. There would definitely be some logistical overhead in labs, but the benefit is so enormous that it's surely worth overcoming-it would need to make testing ten times more logistically challenging before you weren't winning!
@@DrAndrewSteele Do you think that a country (I prefer not to name it out of respect for my flag 🇫🇷) having refused the help of veterinary laboratories at the worst of the crisis is ready for this🙄?
The first preprint I found referring to this technique dates from last April. The capacity to test has been one of the arguments put forward almost everywhere to show muscles. A capacity with objectives formulated in figures, presented as the guarantee of a rapid and efficient response to the emergence of clusters. The goal reached (more than one million per week here, including sometimes one week before getting a result and a failure to respond effectively to the emergence of clusters for more than two and a half months), who knows : perhaps we will now be able to ask ourselves the question of the method...with an average positivity rate of 13 now...
There's no greater risk of producing false negatives or false positives. We had several months to think about it in a context where the positivity rate was less than 3. A latency that have severely testing our laboratories. I don't know how to answer the question of whether if a combinatorial pooling strategy would have posed greater logistical issues. Nor do I know if the one who affixed the motto "Why make it simple when we can make it complicated?" at the pediment of the entrance to administration schools used to go through his bathroom window to open the door from inside every time he was going home.
Thank you Andrew for this uplifting and perfectly substantiated video !
@@anne-sophiej.9584 Thanks! And, er, the UK refused the help of academic biology labs early in the crisis too…I think there may be a lot of similarities…
Nice approach.. seems kinda obvious that it should be implemented. But what about false negatives? The impact of false negatives would have much greater impact using this methods. Still though, you could run 2 of these tests a week apart and use less tests than the default case.
binary search is so amazing 🔥 Still many countries are not permitting for the pool testing :(
I couldn't find a totality on his T-shirt 😅. I want this T-shirt 😍😍
Came from Steve Mould. Subbed but this could get confusing what with being a big Alec Steele fan
Love the maths. They are doing this to a certain extent in the UK, testing students and sticking all the swabs into the same sample from the class. If a class comes back positive the hole class is tested again, indevidualy.
Steve Mould helped me to reach here thanks to him and hi to you man❤️
All of this makes me think of the recent videos on error correcting codes by 3Blue1Brown and Matt Parker. You could pool your tests using the same groups as calculating parity bits in a hamming code, and you could use 6 tests to uniquely identify a single positive in 57 tests
There's also a lot of research on multi-bit error detection/correction that could be used to make this more robust to outlier samples where like 5% are positive.
Idea:
Get as much data of each patient as possible: has been in contact,? when?, has symptoms?, how severe?.
Use neural networks or plain statistics to estimate the chance of that patient of been positive.
Test them in groups of size inversely proportional to their chance of been positive.
If a big group tests positive subdivide it in smaller groups and test those.
Use the results to update the neural networks or statistics.
You can also test wastewaters to estimate the baseline chance of a region.
What does your shirt say?
Great vid,
Your audio is veru quite, even at max volume I have trouble hearing you
Thanks for the feedback! I agree, and will normalise to a higher level on future videos…
How does test false positive and false negatives rates of the test influence pooling tests rates?
See the FAQ in my pinned comment!
cheers steve mould
There’s a strong connection between this and error correcting codes used to make digital transmissions error correcting and error detecting. 3blue1brown’s recent videos on the topic do a great job speaking how the technique works.
It's super interesting. But when talking about testing everyone in the US, number of tests needed aside, don't you make an assumption that corona virus cases are uniformly distributed geographically, or that you're able to mix samples from wider an area of land than is feasible (to be a practically possible strategy)?
Great video. Brought here by Steve's tweet. Also China used the pooling test again to test the entire Qingdao city's population within 5 days. www.google.com/amp/s/www.bbc.com/news/amp/world-asia-54504785. Although they only used the simple pooling instead of the fancy multidimensional pools or combinatorics as you mentioned in this video, still cool to see it in real practice. I suppose there is also a careful balance that needs to be struck between multidimensional pooling methods and the risk of human error being made when the testing pools become less and less intuitive as we go up in dimensions.
Clever, you work for a ThinkTank?
Unnatural stabilization
maybe thats a lack of my english but we just talk about the laboratory test but everybody gets his own teststick in their nose?^^
When you say save 9 test kits, you mean the PCR kit and not the actually swabs right? As everyone still would need to use a "test kit" to begin with
Amazing the simple way the math is explained for it. But unfortunately the testing part of running test in the whole population is the simplest logistic part. Collection, distribution, tracking and contact tracing are the real hard problems to solve. Those are the ones that can't be solved with brute force, more machines, and robots....
Steve Mould did not send me here